mlsanityData to 14 Oct 2026

How Low Can We Go? Pixel Annotation for Semantic Segmentation

arXiv 2201.10448 ↗First posted 25 Jan 2022cs.CVcited by 0

Not on the map. Why?

Abstract (arXiv)

How many labeled pixels are needed to segment an image, without any prior knowledge? We conduct an experiment to answer this question. In our experiment, an Oracle is using Active Learning to train a network from scratch. The Oracle has access to the entire label map of the image, but the goal is to reveal as little pixel labels to the network as possible. We find that, on average, the Oracle needs to reveal (i.e., annotate) less than 0.1% of the pixels in order to train a network. The network can then label all pixels in the image at an accuracy of more than 98%. Based on this single-image-annotation experiment, we design an experiment to quickly annotate an entire data set. In the data set level experiment the Oracle trains a new network for each image from scratch. The network can then be used to create pseudo-labels, which are the network predicted labels of the unlabeled pixels, for the entire image. Only then, a data set level network is trained from scratch on all the pseudo-labeled images at once. We repeat both image level and data set level experiments on two, very different, real-world data sets, and find that it is possible to reach the performance of a fully annotated data set using a fraction of the annotation cost.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 24 of its 44 references that are in the map. First posted on arXiv 25 Jan 2022.

Builds on

24 of 44 references.

References · 24

  1. 2016The Cityscapes Dataset for Semantic Urban Scene Understandingcited by 4,444
  2. 2018Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentationcited by 3,223
  3. 2016Pyramid Scene Parsing Networkcited by 2,479
  4. 2019Self-training with Noisy Student improves ImageNet classificationcited by 1,311
  5. 2017Deep Bayesian Active Learning with Image Datacited by 819
  6. 2020ResNeSt: Split-Attention Networkscited by 445
  7. 2018Rethinking ImageNet Pre-trainingcited by 377
  8. 2020Rethinking Pre-training and Self-trainingcited by 352
  9. 2016ScribbleSup: Scribble-Supervised Convolutional Networks for Semantic Segmentationcited by 331
  10. 2015What's the Point: Semantic Segmentation with Point Supervisioncited by 298
  11. 2019Zero-Shot Semantic Segmentationcited by 245
  12. 2016Deep Interactive Object Selectioncited by 160
12 more
  1. 2018On Regularized Losses for Weakly-supervised CNN Segmentationcited by 121
  2. 2018"Double-DIP": Unsupervised Image Decomposition via Coupled Deep-Image-Priorscited by 114
  3. 2018Normalized Cut Loss for Weakly-supervised CNN Segmentationcited by 112
  4. 2018Learning random-walk label propagation for weakly-supervised semantic segmentationcited by 101
  5. 2019ViewAL: Active Learning with Viewpoint Entropy for Semantic Segmentationcited by 96
  6. 2020Reinforced active learning for image segmentationcited by 56
  7. 2017Not All Pixels Are Equal: Difficulty-aware Semantic Segmentation via Deep Layer Cascadecited by 53
  8. 2019Semantic Redundancies in Image-Classification Datasets: The 10% You Don't Needcited by 43
  9. 2021A Comprehensive Study on Colorectal Polyp Segmentation with ResUNet++, Conditional Random Field and Test-Time Augmentationcited by 41
  10. 2018Interactive Full Image Segmentation by Considering All Regions Jointlycited by 34
  11. 2018Are All Training Examples Created Equal? An Empirical Studycited by 16
  12. 2019Budget-aware Semi-Supervised Semantic and Instance Segmentationcited by 12

Built on by

Not cited yet.